A curated list of frameworks, libraries, tools, platforms, protocols, and resources for building agentic AI systems. Last updated: February 2026.
Agentic AI refers to AI systems that can autonomously plan, reason, use tools, and complete complex multi-step tasks with minimal human intervention.
| Version | Date | Notes |
|---|---|---|
| v1.0.2 | Feb 2026 | Added VoltAgent to Frameworks & Orchestration and awesome-ai-agent-papers to Other Awesome Lists after reviewing the VoltAgent GitHub organization. |
| v1.0.1 | Feb 2026 | Reviewed against GitHub awesome-ai topic page; added missing relevant lists in Other Awesome Lists (awesome-ai-coding-tools, awesome-claude, awesome-github-copilot, awesome-ai-agents). |
| v1.0.0 | Feb 2026 | Initial release. Covers frameworks, multi-agent systems, coding agents, protocols, LLM models, memory/RAG, tool use, observability, low-code builders, local tooling, cloud platforms, benchmarks, research, and learning resources. |
- Changelog
- Frameworks & Orchestration
- Multi-Agent Systems
- Coding Agents
- Protocols & Standards
- LLM Models for Agents
- Memory & Retrieval (RAG)
- Tool Use & Function Calling
- Observability & Evaluation
- Visual / Low-Code Builders
- Local LLM Tooling
- Cloud & Enterprise Platforms
- Benchmarks & Datasets
- Research & Papers
- Learning Resources
- Other Awesome Lists
General-purpose frameworks for building agentic LLM applications.
| Name | Stars | Description |
|---|---|---|
| LangChain | 123k+ | The OG framework for chaining LLM calls, tool use, memory, and RAG pipelines. |
| LangGraph | 25k+ | Stateful, graph-based orchestration for complex multi-step agent workflows. Built on LangChain. |
| LlamaIndex | 46k+ | Framework for connecting LLMs to data sources; strong RAG and agent workflows. |
| Pydantic AI | 12k+ | Type-safe, validation-first agent framework from the Pydantic team. Ideal for compliance-heavy use cases. |
| Haystack | 22k+ | End-to-end NLP framework with pipeline-based agent orchestration from deepset. |
| DSPy | 24k+ | Stanford's framework for programming LLM pipelines using declarative modules and auto-optimization. |
| smolagents | 25k+ | Hugging Face's minimal, code-first agent framework (~1k lines of core logic). Model- and tool-agnostic. |
| Mastra | 6k+ | TypeScript-native agentic framework with workflows, memory, and tool support. |
| VoltAgent | 6k+ | Open-source TypeScript AI agent framework with workflows, memory, MCP, RAG, guardrails, and production-oriented ops integrations. |
| Agno | 22k+ | Production-grade multi-agent framework with strong memory, context, and documentation. (formerly phidata) |
| Semantic Kernel | 26k+ | Microsoft's SDK for integrating LLMs into C#, Python, and Java apps with agentic patterns. |
| Instructor | 10k+ | Structured outputs from LLMs via Pydantic schemas. Pairs well with any agent framework. |
Frameworks specifically designed for coordinating multiple collaborating agents.
| Name | Stars | Description |
|---|---|---|
| AutoGen | 53k+ | Microsoft's multi-agent orchestration platform for collaborative, conversational agents. |
| CrewAI | 42k+ | Role-based multi-agent teams with explicit crew definitions and task delegation. |
| OpenAI Agents SDK | 10k+ | OpenAI's production-ready multi-agent SDK with built-in tracing, guardrails, and memory. Replaces the deprecated Swarm. |
| AutoGPT | 177k+ | The original autonomous agent platform. Goal-driven, self-prompting LLM agent. |
| AgentGPT | 32k+ | Browser-based, no-code autonomous agent builder. |
| BabyAGI | 20k+ | Minimal, educational task-driven autonomous agent for research and experimentation. |
| MetaGPT | 48k+ | Multi-agent framework that assigns GPT agents software engineering roles (PM, dev, QA). |
| ChatDev | 27k+ | Simulates a software company with LLM agents playing different development roles. |
Agents and tools specifically designed for autonomous software engineering tasks.
| Name | Stars | Description |
|---|---|---|
| OpenHands | 50k+ | (formerly OpenDevin) Open-source, modular platform for autonomous software agents. MIT licensed. |
| SWE-agent | 14k+ | Princeton's agent for autonomously resolving GitHub issues. Powers the SWE-bench benchmark. |
| Open Interpreter | 60k+ | Lets LLMs run code locally. Natural language interface to your computer's capabilities. |
| Aider | 24k+ | AI pair programmer in your terminal. Works with Git and supports most major LLMs. |
| Cline | 20k+ | Autonomous coding agent for VS Code with MCP support and browser use. |
| Cursor | — | AI-powered IDE with agentic code editing, Composer, and multi-file context. |
| Windsurf | — | Codeium's agentic IDE with Cascade flow engine for collaborative coding. |
| Devin | — | Cognition AI's fully autonomous software engineer agent (commercial). |
| GitHub Copilot | — | GitHub's AI coding assistant, now with agentic coding mode in IDEs. |
Open protocols enabling interoperability between agents and tools.
| Name | Origin | Purpose |
|---|---|---|
| Model Context Protocol (MCP) | Anthropic | Universal standard for connecting LLMs/agents to external tools, APIs, and data sources. Vertical (agent ↔ tool). |
| Agent-to-Agent (A2A) | Peer-to-peer protocol for agents to coordinate, delegate, and collaborate across vendors. Horizontal (agent ↔ agent). | |
| Agent Communication Protocol (ACP) | IBM / BeeAI | RESTful, session-aware protocol for structured agent message exchange. Internal comms. |
| Agent Network Protocol (ANP) | Community | Decentralized agent discovery and collaboration using W3C DIDs for open networks. |
Quick guide: Use MCP for tool access, A2A for cross-agent teamwork, ACP for intra-cluster messaging, ANP for decentralized agent networks.
Frontier and open-source models well-suited for agentic workloads (tool use, long-horizon reasoning).
| Model | Provider | Notes |
|---|---|---|
| GPT-5.2 | OpenAI | Current flagship model. Supersedes GPT-4o/o3 (retired Feb 2026). Best-in-class tool calling and reasoning. |
| Claude Sonnet 4.6 / Opus 4.6 | Anthropic | Latest Claude generation (Feb 2026). 1M token context, superior agentic coding and extended thinking. |
| Gemini 3 Pro / 2.5 Flash | Gemini 3 Pro is Google's latest flagship; 2.5 Flash remains available for high-speed, multimodal agents. | |
| Grok-3 | xAI | Strong coding and reasoning capabilities. |
| Model | GitHub | Notes |
|---|---|---|
| DeepSeek-V3 / R1 | deepseek-ai | State-of-the-art open model; cost-effective for agentic workflows at scale. |
| Qwen 3 | Alibaba / QwenLM | Latest generation; top-tier multilingual, math, and code-capable open model (Apache 2.0). |
| Mistral / Mixtral | Mistral AI | Efficient MoE models; good balance of quality and cost. |
| Llama 4 | Meta | Meta's latest open family with 10M token context; strong reasoning and coding for agentic use. |
| Phi-4 | Microsoft | Small but powerful; great for edge/local agent deployments. |
| Kimi K2 | Moonshot AI | High-performing open model for tool use and agentic action completion. |
Libraries and services for giving agents persistent memory and knowledge retrieval.
| Name | Stars | Description |
|---|---|---|
| Mem0 | 28k+ | Intelligent memory layer for AI agents. Stores, updates, and retrieves memories across sessions. |
| LangMem | 2k+ | LangChain's native long-term memory SDK for agent state across conversations. |
| Chroma | 18k+ | Open-source, developer-friendly vector database for embeddings and RAG. |
| Weaviate | 13k+ | Open-source vector search engine with hybrid (vector + keyword) search. |
| Qdrant | 22k+ | High-performance vector similarity search, written in Rust. |
| Milvus | 33k+ | Scalable, cloud-native vector database built for billion-scale embeddings. |
| pgvector | 14k+ | Open-source vector extension for PostgreSQL. Simple and production-ready. |
| Zep | 3k+ | Long-term memory store and knowledge graph for LLM agents. |
Libraries that help agents discover, call, and manage external tools and APIs.
| Name | Stars | Description |
|---|---|---|
| Composio | 17k+ | 250+ pre-built tool integrations (GitHub, Slack, Gmail, etc.) for AI agents. |
| Browserbase | 1k+ | Cloud browser infrastructure for headless browser automation in agents. |
| Browser Use | 60k+ | Library enabling LLM agents to control and interact with web browsers. |
| Playwright | 70k+ | Microsoft's browser automation library widely used for agent web interactions. |
| Toolhouse | 1k+ | Cloud-hosted tool execution layer for agents with built-in auth and caching. |
Monitor, trace, debug, and evaluate your agent pipelines in development and production.
| Name | Stars | Open Source | Description |
|---|---|---|---|
| Langfuse | 10k+ | ✅ | Deep LLM tracing and observability; self-hostable. Integrates with most frameworks. |
| LangSmith | — | ❌ | LangChain's hosted tracing, debugging, and dataset curation platform. Minimal overhead. |
| Arize Phoenix | 5k+ | ✅ | OpenTelemetry-native agent tracing and evaluation for enterprise ML teams. |
| Weave | 2k+ | ✅ | W&B's tool for tracking, evaluating, and iterating on LLM/agent pipelines. |
| Helicone | 3k+ | ✅ | Lightweight API monitoring for token cost, latency, and caching. |
| AgentOps | 3k+ | ✅ | Agent session replay, cost tracking, and error detection. |
| Braintrust | — | ❌ | Enterprise eval platform with fine-grained scoring for agent outputs. |
| Maxim AI | — | ❌ | End-to-end simulation, evaluation, and observability for agentic systems. |
| RAGAS | 8k+ | ✅ | Evaluation framework specifically for RAG pipelines—faithfulness, relevance, etc. |
Build agent workflows visually without deep coding required.
| Name | Stars | Description |
|---|---|---|
| Langflow | 55k+ | Drag-and-drop visual builder for LangChain-based agent pipelines. |
| Flowise | 40k+ | No-code LLM flow builder with prebuilt nodes for chains, agents, and tools. |
| Dify | 80k+ | Open-source LLM app development platform with prompt IDE, RAG, and agent tools. |
| n8n | 90k+ | Workflow automation tool with native AI/LLM nodes for agentic task pipelines. |
| Rivet | 3k+ | Open-source AI agent runtime and visual programming environment by Ironclad. |
| PromptFlow | 10k+ | Microsoft's toolkit for building, evaluating, and deploying LLM flows. |
Run and manage LLMs locally for private, offline, or resource-constrained agent deployments.
| Name | Stars | Description |
|---|---|---|
| Ollama | 165k+ | The easiest way to run LLMs locally on Mac, Windows, and Linux. |
| LM Studio | — | Desktop app for discovering, downloading, and running local LLMs with a GUI. |
| vLLM | 50k+ | Fast, memory-efficient LLM inference server. Production-ready local hosting. |
| llama.cpp | 75k+ | C/C++ inference for LLaMA models; runs on CPU and GPU with quantization. |
| GPT4All | 72k+ | Open-source ecosystem for running powerful LLMs locally on consumer hardware. |
| Jan | 26k+ | Open-source offline-first personal AI assistant with local model support. |
Managed platforms and cloud offerings for deploying production agentic AI systems.
| Name | Provider | Description |
|---|---|---|
| Amazon Bedrock Agents | AWS | Fully managed agent builder with tool use, knowledge bases, and guardrails. |
| Azure AI Agent Service | Microsoft | Managed agents on Azure with AutoGen orchestration and enterprise security. |
| Google Vertex AI Agents | Google Cloud | Build, deploy, and manage agents using Gemini models on GCP. |
| Salesforce Agentforce | Salesforce | Enterprise autonomous agents for CRM, customer service, and business workflows. |
| ServiceNow AI Agents | ServiceNow | Domain-specific agents for IT, HR, and customer support automation. |
| Relevance AI | Relevance AI | Visual multi-agent workflow orchestration for enterprise teams. |
| Cognitiev / E2B | E2B | Sandboxed code execution infrastructure for running agent-generated code safely. |
Evaluate agent capabilities on standardized tasks.
| Name | Description |
|---|---|
| SWE-bench | Evaluates agents on resolving real GitHub issues across popular Python repos. |
| GAIA | General AI Assistants benchmark requiring real-world reasoning and tool use. |
| AgentBench | Multi-environment benchmark covering web, DB, OS, and game environments. |
| WebArena | Realistic web browser environment benchmark for autonomous web agents. |
| TAU-bench | Task-oriented benchmark for evaluating LLM agents in real-world API tool use. |
| BrowsingBench | Evaluates agents on complex multi-step web browsing tasks. |
| OSWorld | Benchmark for desktop GUI agents completing OS-level tasks. |
Seminal papers and surveys shaping agentic AI research.
- ReAct: Synergizing Reasoning and Acting in Language Models — Yao et al. (2022) — Foundational ReAct prompting pattern for agents.
- Toolformer: Language Models Can Teach Themselves to Use Tools — Schick et al. (2023)
- HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in HuggingFace — Shen et al. (2023)
- Voyager: An Open-Ended Embodied Agent with Large Language Models — Wang et al. (2023)
- AgentBench: Evaluating LLMs as Agents — Liu et al. (2023)
- OpenHands: An Open Platform for AI Software Developers as Generalist Agents — Wang et al. (2024)
- A Survey of Agent Interoperability Protocols: MCP, A2A, ACP, ANP — 2025 survey
- Agentic AI Frameworks: Architectures, Protocols, and Design Challenges — 2025 survey
- DeepLearning.AI: AI Agents in LangGraph — Hands-on short course.
- DeepLearning.AI: Multi AI Agent Systems with CrewAI
- DeepLearning.AI: Building Agentic RAG with LlamaIndex
- Hugging Face: Agents Course — Free, open-source course on building AI agents.
- LangChain Academy: Introduction to LangGraph
- Lilian Weng's Blog — Deeply technical posts on LLM agents and planning.
- Simon Willison's Weblog — Practical notes on LLMs, tools, and agents.
- The Batch (deeplearning.ai) — Weekly AI newsletter.
- DAIR.AI Newsletter — Top ML papers weekly.
- Ahead of AI — Sebastian Raschka's newsletter on LLMs and agents.
- Andrej Karpathy — Deep dives into LLMs and neural networks.
- AI Jason — Practical tutorials on AI agents and tools.
- Sam Witteveen — LangChain, LLM agents, and RAG tutorials.
- awesome-ai-coding-tools — Curated list of AI-powered coding tools.
- awesome-claude — Curated list of Claude ecosystem tools and resources (including coding-agent workflows).
- awesome-github-copilot — Prompts, agents, skills, and practical resources for GitHub Copilot.
- awesome-ai-agents (NipunaRanasinghe) — Curated frameworks, tools, and resources for AI agents.
- awesome-ai-agent-papers — Curated collection of recent AI agent research papers (agent engineering, memory, evals, workflows).
- awesome-ai-agents — 300+ agentic AI projects with metrics and categories.
- awesome-llm-agents — Curated LLM-powered agent papers and projects.
- awesome-langchain — Tools and projects built with LangChain.
- awesome-mcp-servers — Curated list of MCP servers for Claude and other agents.
- awesome-chatgpt-prompts — Prompt engineering resource for LLMs.
- LLM-Agents-Papers — Comprehensive list of agent research papers.
- Awesome AI Agents (korchasa) — Auto-maintained directory of multi-agent systems.
Pull requests are welcome! Please ensure entries are:
- Actively maintained (last commit within 12 months, or historically significant)
- Relevant to agentic AI (autonomous decision-making, tool use, planning, multi-agent)
- Accompanied by a clear, one-line description
This list is released under the CC0 1.0 Universal public domain dedication.
Inspired by the awesome list format.